Name: IBM’s Watson Will Tackle Cancer Next
Uploaded: 2016-11-02
Duration: 12 min 59 s
Description: The company’s VP of oncology shows us how

[ MUSIC ] When it comes to democratizing knowledge, this is one of our prime focus areas in Watson. And it's not about exporting knowledge around the world, you know, quite frankly some of the institutions you mentioned are Perfect users of this capability. Now, there's been a lot of wonderful discussion the past two days on disruptive technologies as well as data proliferation. And I'm very respectful of this group, but I think it's worth providing a little bit of context before we jump into the demonstration. And so, related to information, related to knowledge and the proliferation of knowledge Think about the challenge for any organization within healthcare. We used to say that medical literature doubled every five years. Then we said medical literature doubled every three years. I think we're up to two years and by 2020 in several months medical literature will double. So no matter How wonderful of an institution you're in, or how great of a reader you are. You couldn't possibly have the capacity to keep up with that information. And we can't keep up with our e- mails. Exactly, and that's before you get to things like clinical trials, were there is 180, 000 clinical trials in the NIH database at any given point in time, or research in a million articles per year And so now the beauty of having cognitive computing this new era of technology. You have a mechanism for analyzing these large information sources and leveraging to get exactly the incite you need at exactly when it's required. So within [ UNKNOWN ] we have several solutions that Assist across the clinical continuum. Yesterday, we mentioned we have solutions that can analyze medical images. Mammograms being one of the first use cases. In addition to that, we have solutions that can assist molecular tumor review board analyze genomic considerations for a patient within ten minutes. And so any of you who are familiar with that process know what a substantial decompression in time that is and what an efficiency that is for very, very talented people that are trying to analyze that information. We have solutions that can analyze clinical trials at the point of care so that you can immediately see what a patient is eligible for. Bottoms up, based upon the trial protocol. And the solution that I'm going to share today is focused on approved treatment options. So imagine now you can have Watson assist, as you mentioned, by democratizing knowledge on what the appropriate treatment options are. Now I heard a little bit of debate over the last day, as far as, what percentage that would be, that specialists tend to know these things. But I'll just Give you a data point related to lung cancer. One of the physicians that we worked with who trained Watson in lung cancer said when he started his career there were essentially four primary interventions for lung cancer. Now there's over 600. And that's before you get into precision medicine so even though the best of the best the sub specialists can benefit from capabilities like this to analyse information. So if we can bring Watson onto stage with us I can share with you a demonstration into the capability. Okay so what you see here are patient cases and I prepared a case for today That is basically a stage 2b breast cancer patient. And the way this works is, imagine Watson is integrated with the electronic health record. And hundreds of notes, whether they be nurse encounters, physician encounters, lab notes, are fed into Watson. And Watson reads it. So the technologists in the room, it could be a continuity of care document. Mm- hm. It could be the equivalent of a Word document that's a lab report. And just as Watson demonstrated on Jeopardy five years ago that it understood human language. Now Watson understands what tumors are and what stages of tumors are. So just to bring this into human language, a Stage 2D breast cancer patient is a woman with breast cancer that's spread regionally? Know that it has, well, there's different mechanics for it. Okay. But let's see what Watson will tell you. Okay. So what you're seeing now is a summary of this patient and so Watson basically has information on this individual, as you can see that it derived from the medical records. So it's a 62 stage 2B breast cancer patient and In here should be various pieces of the status, so she's Positive, PR positive, and I was looking for- Hormone receptors. Exactly, hormone receptors, and then you could see there's things like performance status Where there's different measurements for this, but basically this person's asymptomatic and in this example, Watson found that the Karnofsky score, another measurement for performance status, was included in one of the notes and prepopulated this field. And I'm just trying to find the size of the tumor Which I can probably just get here. [ BLANK_AUDIO ] Warning, you're close to the stage here. Yeah, no, no, no. That's fine, I just lost one of the fields but that's okay. So basically it's a 7. 4 cm tumor. Okay. And that's basically what drives it into a steady [ UNKNOWN ] Phase two. Exactly. So now, the first thing a physical will do is validate all the information for this patient, and once it's verified, then what Watson can do is bring up the various treatment options. So Watson isn't winging it, right? The physician is part of this process. No, so now, what's happening is In addition to Watson reading through the electronic health record and figuring out what's relevant for this patient in the context of cancer care, what Watson also has been loaded with is 15 million pages of medical literature and it's been trained on thousands of cases That have been curated so Watson can understand the right treatment for these individuals. And so as it went through this process, it basically brought up the treatment options now specific to this individual and there's essentially three sets of options. There's recommended, generally good in any circumstance, that's green. There's yellow which is for consideration, could very well a good option, but there may be something you need to explore like a [ UNKNOWN ] just as one example or red generally not recommended. Could be a [ UNKNOWN ] a treatment option and so the first thing you see on the top here is the treatment plan Which is basically recommending a [ UNKNOWN ] chemotherapy, followed by surgery, endocrine and radiation, and you can see the overall timeline for the patient. And then what you see below are the specific treatments options within that timeline. And so what you can do here, as an example, is click on the various option And now imagine I'm using a laptop because it just works better with the screen resolution. Imagine you have a tablet and you're a physician in front of a patient. Ad up comes all of the literature supporting why this is a good treatment option specifically for this patient. And so you can see things here like survival statistics for this patient. So those numbers that you're showing are the survival percentages of patients that survive. With this treatment option by year. This treatment. Right. And then you can It's the other things like toxicity statistics and then if you just want to say, hey, I want to see what other literature may be out there? There's a button you could click where the latest publications will be pulled up relevant to this patient You can click on Administration, so you get information like dosing. And you can also then get other information related to the drugs, so you can either look at as an example at the entire drug label, or it could be personalized. You could say only show me The side effects that may be specific to this patient, and Watson can bring up both, and do it in a very novel way. So, I'll just click on one of the chemotherapies here, and up comes the very various adverse events by incident percentage, as well as highlighting what's severe or moderate. And so now imagine Again as a patient, in addition to having the physician informed on the evidence supporting the various treatment options, he or she can then turn around the iPad, show you all this information, So you can take a shared decision together on what's right for you factoring lifestyle considerations and things like that And then at the end of the process, what happens is you can print off all of this information, you can push it back into your electronic health record so that you can retain what was decided and all the evidence supporting that treatment option. So, in a moment, we have a few minutes for questions and I definitely want to see some hands here but One quick question is who can get this, and how much does it cost? So basically, it's operating in hospitals around the world. One of the announcements, a couple of weeks ago, as an example, is IBM has made this available to our employees, so there's a second opinion service now that can leverage it. So it's not something that we go direct to consumer with. It's something that is a partnership with the clinician, so we sell it to various physicians across the Watson Health portfolio. We're working with probably 100 hospitals right now using various technologies. Okay, questions, anybody? Yeah, right there. [ BLANK_AUDIO ] So this is very fascinating. I just have a question around the integrity of the information that you're aggregating. I'm asking because it says that you're focused on approved products, and what we know is that only 56 % of the clinical trial results make it into the medical literature. And only 20 percent make it into that NIH clinical registry that you were talking about. So how are you handling the partial information that you have? And then the other piece is How are you getting around the pay walls for the medical journals? Are you using just the abstract or what's the mechanism? So, great, great questions. So first of all there's a portfolio of offerings that come together in a platform, if you will, in the context of oncology. And so what we are showing is one of those solutions related to approved treatment options. So all the literature in here is basically peer- reviewed literature. And what we've done is we've segregated it into curated literature versus just all other literature. Point one. Related to trials, you don't have to limit it to the NIHClinicalTrials. gov. What we're doing with many of our clients is actually loading in their proprietary protocols, so that they can apply that against their patient cases as well. And so there's a hospital system who just about every breast cancer patient goes through this technology. And they don't only limit it to clinicaltrials. gov, they look at their proprietary protocols. And then there's another part of your question that I didn't answer. [ CROSSTALK ] No, through the paywall. All these publishing companies charge a lot just to access their literature. Do you just read the abstracts or do you actually get the entire long. It depends on the solution. So IBM can't afford the medical literature, either, for those. [ LAUGH ] Nobody can afford it. So in this solution, we have 15 million pages of medical literature and our genomic solution, which Which has been getting a lot of press, it's probably closer to 40 million so it depends on the solution and what level of detail you put in. Alright. We have one minute left, does anyone have a five second question that we can answer in fifty seconds. Anybody. Okay. I will ask you so in terms of the training, just really quickly, when you talk about training how do you train the computer. Well it's interesting and the analogy I always use with people. Even technologists, as I said, let's not lose sight of the fact that what we're trying to do is basically, replicate human brain function. And so the training is very similar. What you saw on Jeopardy is Watson went to school and learned nouns and verbs and then ingested many, many knowledge sources for factoid trivia. We then sent Watson to medical school, where it learned the vocabulary and [ UNKNOWN ]- That took, how long that took? Well the first few diseases. Now there's some research involved and it took about three years, but the solution- Same, medical school. Exactly, the solution we can train about disease per month, and that includes adapting it to a specific institution's pathway if you would like to customize it for your organization. And so it's a combination of ingesting literature, using training cases and then testing it with subject matter experts, just like you would in a university. Thank you. [ MUSIC ]

The company’s VP of oncology shows us how