- IBM’s Watson supercomputer has cemented its place in game show legend for its stunning triumphs on Jeopardy!, but answering trivia questions was just the beginning of a long and fruitful legacy for this data-crunching powerhouse. Putting aside the fun and games, Watson has turned its attention to some serious matters: cancer research, medical exams, and bringing its novel style of natural language processing to bear on clinical decision support for providers and concierge medicine for consumers.
At HIMSS14 in February, Steve Gold, vice president of IBM’s Watson Group, spoke to EHRintelligence about how cognitive computing is opening up new opportunities for providers, consumers, and researchers to harness some of the most advanced capabilities Big Blue has to offer.
What is so unique about the way Watson consumes data and returns answers to the user?
Jeopardy! was a very public prove point that really showcased the fact that a computer can do something that had never been done before. We’re so used to the idea of computational computing. Everything is programs. It’s based on rules and logic and structured data, and that’s what we’ve come to accept. But Watson’s really good at collecting information and understanding it in a different way. Watson’s really good at natural language, and ingesting vast and disparate forms of data and understanding it in context. And it’s learning.
You want the system to get progressively smarter. If patients are continually being referred for MRIs for back pain, and the evidence is starting to indicate that that’s not actually the appropriate course of treatment, you don’t want to persist in that fashion. You want the system to get smarter, and Watson can do that. It’s kind of like a kid. Second grade is tough, because you’ve got to learn the fundamentals. But then you progress to third grade and fourth grade. By the time you’re at fifth grade, you can start to ask questions that you haven’t necessarily been trained on, and use the learning of second, third and fourth grades to build the reasoning, and that’s what Watson’s doing now.
What has Watson been working on in the healthcare field that has really shown promise?
One of the things we’re doing at Cleveland Clinic is working with medical students, because it turns out that studying for the US Medical License Exam is situational in nature. If you’re a student, they give you questions about how a patient presents. “A 41-year-old female presents with the following symptoms, so what would you do?” Want to guess the average proficiency necessary to pass the U.S. Medical License Exam? Sixty to sixty-five percent. Now, I seem to recall that translates to a D. And I’m thinking, that’s not the grade I want my physician to come out with, right?
Watson’s really good at that type of integration. It can go against volumes of information. Helping doctors navigate the massive amount of information is so critical to improving the quality of care that they can then deliver. And then in October, we announced that we were working with M.D. Anderson to really bridge the divide between the research side of the house and the clinical side of the house.
If you have cancer, what you really want to do is get into a clinical trial. You don’t want conventional treatment. But getting matched to a clinical trial is really hard. Much harder than I ever envisioned. M.D. Anderson runs hundreds at any one point, and trying to figure out the criteria for clinical trial against the patient circumstance, comorbidities and attributes and family history and medication is really hard. Turns out, Watson’s really good at finding needles in haystack patterns, and it’s been very effective for patient matching like that.
How will Watson interact with users on a consumer level?
The vision is to use Watson to help people keep New Year’s resolutions, as I like to say. So, my New Year’s resolutions are that I’m going to eat healthy and I’m going to exercise more. Right. By January 15, that’s over, isn’t it?
But why is that? Well, because I don’t know which foods are healthy. I don’t know what exercise I can do at my desk, for example, and there’s really no one to ask. The vision is to ask Watson. So let’s say I just came from my doctor, and my doctor said I have high cholesterol. What can I do? All right, he’s prescribing medication X. What is that? Some statin, but what are statins? Maybe he didn’t explain it properly. Maybe I don’t know where to do my own research.
Or maybe I go home after a hospital procedure, but I’m feeling fuzzy. They handed me these discharge papers. I have no idea what they say. Hours later, I’m going to take my medication. Well, this one says take every four hours with food, but this one says every six hours on an empty stomach. Do I take them two hours apart? What do I exactly do I do?
This notion of a healthcare concierge or a personal coach, if you will, is going to be very appealing to the users. And today, that represents 17 million members. So we’re thinking this whole ecosystem is kind of neat. You’ve got Watson being applied to the way medicine is being taught, researched, practiced and paid, and it’s being consumed by professionals and by consumers. So, we truly believe this is a transformational play.
How will providers benefit from Watson’s clinical smarts?
When we talk to providers, one of the comments we hear consistently is the frustration over the time it takes to interact with their technology. And if you’ve been to the doctor recently, you know that they spend 80% of the time typing into the EHR and 20% of the time actually talking to you as the patient. And unfortunately, for a lot of reasons – policy reasons, process reasons, regulatory and recording requirements – they have to do it. What they’ve said to us is, “I don’t want to change my process, but I want to find a way to move through the process more seamlessly.”
Natural language is, to them, far more intuitive. The ability to navigate through speech or through text without having to decipher a user interface is highly attractive. It does not displace the fact that the EHR exists. This just becomes a way into the EHR that really affords them, I think, greater time with the patient.
The other problem that they cite is that when they walk into the examination room, the first thing they do is they pull up the EHR and start reading the patient’s history, patient reports, following symptoms, patient history; maybe they go through the medications list. And then they’re trying to do an assessment. Watson pulls that all together for them. Watson presents what it believes to be the relevant information in a way that’s quickly ingestible and understandable. So they don’t spend as much time on the front-end doing reading, which is good for them and good for the patient. They don’t spend as much time on the back-end inputting, which is good for them and the patient.
We know that evidence-based medicine is the best way to practice. Today, if a physician wants to reference something, there’s a really good chance he would actually walk over to a bookshelf, pull one of the reference books, flip to page 743, find the paragraph in the passage, read it, and feel more confident in his decision. Not exactly efficient, but that’s what he would do. But everyone knows that getting to the data, discovering it, uncovering it, presenting it in the eight or ten minutes that I am afforded with that patient…it’s close to impossible.
Watson can help provide that assistance. It’s not making the decision for the doctor. It’s not making the decision for the patient. It’s providing a set of possible actions that the doctor and the patient will ultimately use to make a final determination. And that’s actually what is happening.
We’re doing this at Memorial Sloan-Kettering with cancer treatments. Five years ago, it wasn’t very hard for an oncologist to dictate the course of treatment for a condition. Today, it’s really challenging due to the combination and permutations of different chemotherapies and radiation and surgical options. It’s difficult to decipher.
All Watson is doing is saying, “Look, based on the patient profile, family history, medical evidence, Treatment Plan 1 has the highest confidence associated with the best possible outcome based on known information.” Now, for various reasons, the physician and patient may choose Plan 2. Option 1 may be 76% confidence that’s the right option. Option 2 may be 72% confidence. It doesn’t prevent the physician and patient from choosing what they think is right, but it presents all the information.
What’s next for Watson?
Scale. In a word, scale. We’ve moved from this notion of a novelty at a game show to establishing that there is a commercial component to this that’s very viable. As a consumer, today, you can’t touch Watson. By the end of the year, you’ll be working with Watson. So, you’ll start to see it in more consumer-oriented.
We are in the early days of the development of cognitive applications. We’re going to see new, creative, innovative uses. We’ve been living in the programmatic era since the 1950’s, but I’m firmly convinced in 20, 30, 40 years, we’ll still be talking about cognitive computing. We’re going to see the technology evolve. We’ll see new players come to market. All of that, we think is great. But if you ask me next year, “So, what’s next?” I would probably go, “Scale.”