Imagine a future in which a software program, tapped into a database containing every electronic medical record in the country, helps guide your every interaction with your doctor.
During a long conversation at MassDevice.com‘s Boston offices recently, van Terheyden laid out his vision for the integration of Nuance’s speech recognition, mobile devices and Big Data.
MassDevice.com: How can speech recognition play a part in leveraging "Big Data" to enhance the interaction between doctor and patient?
Nick van Terheyden: We have to create much more useful information, so that requires 2 things. One, the extract of the content into some form of codes and information, in part because clinicians use multiple terms for the same clinical conditions. In essence what I’m focusing on is bringing that form of artificial intelligence to voice enablement of healthcare tools and systems. My pretext is based on the overwhelming tsunami of clinical data we’re faced with.
I spend a fair amount of time talking about what that is and why that’s important and why it’s only going to get worse, particularly if you think about genomics, proteomics, individualized medicine. It’s just exploding even further.
The idea that clinicians can process that information at the point of care in a way that will deliver the best possible care is, at this point, long since gone. That train left the station. A good example of that is oncology. If a patient with cancer goes to see a radiation oncologist, they get irradiated. If they go to see the medical oncologist they get poisoned, and if they go to see the surgical oncologist he cuts it out.
But that may not be the right care. So how do we bring that information that we know in most instances and we can demonstrate historically has not been brought to the point of care? It’s tragic that we know the best form of care for people that have suspected myocardial infarction includes a beta blocker, aspirin, at the point of first touch but that that occurs in less than 50% of patients. Really? Just mind-blowing statistics to me, because if that was my parent, my sister, brother…
"We have the information. It’s not that it’s not available; it’s not accessible.”
To be disruptive, this has to be incorporated into the flow. We have the information. It’s not that it’s not available; it’s not accessible. To me it’s never been a data problem, it’s always been a filtering problem – that’s the challenge that clinicians are facing. The integration of voice allows you to listen in, so there’s some potential to incorporate that understanding as part of the sort of passive process. So that no matter who you’re accessing, there’s an intelligent assistant listening in to the conversation that identifies that you’re talking about the following conditions or possible conditions.
Here’s a good example of this wealth of data. Each hospital and medical center in Boston has a huge database, but they’re locked away in this narrative data set. In that information is data that would help treat patients today, but we can’t get to it unless you actually physically read it. So we can mine that and potentially apply it.
MassDevice.com: Where does the electronic medical record fit in this scenario?
NVT: I think that paper’s done; we have to move off of that. But just digitizing that didn’t help us. You only have to look at any of the interfaces to know that that’s an exercise in frustration for most clinicians. Talk to them and, as I describe it frequently, it’s like washing the dishes. Before the dishwasher, nobody wanted to wash the dishes. Then we got the dishwasher and now nobody wants to load the dishwasher. Same with the electronic medical record; it is too complicated to get it in, so they do the bare minimum.
At least 60% of that information is captured through our existing terms and processes that capture audio’s route through transcription, but it’s bundled as this block.
MassDevice.com: So you have to develop something that can talk to all of the different EMR solutions out there.
NVT: Right. So a couple of approaches or ways of addressing that. One, we’ve created a layer that allows for this integration to be incorporated into these mobile platforms. We now have over 200 that have bought into that since HIMMS in February. The integration process can occur in a prototype form over a 1-hour conference call online. I’ve seen it happen. Very impressive and of course, from a partner standpoint, they are very excited about that because now they make the right calls. There is a little bit of thought in terms of how you do it.
And step 2 is pushing out to those others to say, "How do we get in?" The good news about the government, irrespective of your political views, is that 1 of the components of meaningful use is that you must share data. Meaningful use Stage 2 will incorporate HL7-CDA. Like or hate the standard, it doesn’t matter. It is an open standard that allows us to share information in a way that is effective.
I think it will be incorporated into the standard, I think it’s a fairly certain thing that offers 2 ways to get to that data. One it means that it now should be available. We’ll be at that Stage 2 so we’re talking a couple of years out, so that could be frustrating. But the other is, if we provide a layer to our partners, then they are going to handle the access of that information.
MassDevice.com: Give me a real-world scenario 5 years out for the typical doctor with an iPad or iPhone confronting a case that may be outside of his bailiwick. How would this system help?
NVT: I think potentially realistic is the integration of this technology into mobile platforms, and the interaction I think will be through voice.
The clinical process begins immediately, actually as the patient walks in the door, the physician has a differential just by looking at the patient. Technology is not doing that, but as they are talking, I’m starting to refine that. That’s what is going on behind the scenes. So, as the doctor is entering and capturing that information, it’s being passed out to our intelligence behind the screens for a couple of things. One, it’s excluding not necessarily rarified things, but the things that you need to exclude, and they can be excluded easily. That’s critical, because 1 of the things that keeps clinicians up at night is, "What did I miss?" and more importantly, "What did I miss that if I’d caught it, it would make the difference of significant variation and outcome?"
“Medicine isn’t a single-plane decision tree. It’s really a complex matrix.”
The other challenge I see is how this is integrated. I don’t think speaking back is going to work, because that intrudes in terms of the physician-patient interaction. My concept is something along the little pop-ups you get for the notifications, something that comes up and says, "Hey, ask the following question to rule out…" or, "Don’t forget this."
Medicine isn’t a single-plane decision tree. It’s really a complex matrix. To me it’s like a ball of string, and I’m pulling a single thread and, "Oh, that didn’t work," so I’m pulling and as I do, I’m getting more information. That’s how I envisage it from the 3-dimensional standpoint. So, prompting allows me to exclude quickly, easily without all the complexity of testing. Importantly, anything that I have in terms of information, so that access to all that other data says, "I don’t need to ask for liver function tests if they just had it done 1 month ago with Dr. Rex."
I’ll paint another picture and this really is out there. If I was to really project this out, this is a passive system that listens. The idea that we can’t tell that it’s me talking, not you, is ridiculous. I can go to my Xbox and it has directional microphones, based on visual identification of me using ultrasonics, and then a directional microphone that says, "Hey, I picked up the fallen pieces of information."
Why can’t we do that, access that so we can start to filter? You, the patient, come in and I am interacting with you and we are pulling that information and putting it in too, so I’m not having to interact so much with the technology. It sits there, it’s populating. I’m going to verify it. The patient will verify it, and then all of those other things come in the background. I know that’s a little bit way out there, but I don’t think it’s unreasonable.
And as you become more experienced, it becomes more natural and you gloss over some of those things because it’s appropriate. Somebody comes in with an earache, you know you’ve got to exclude a couple of things but you don’t need full neurological exams. But the technology is doing all that and helping you make sure that you don’t miss something because you are on your 50th patient or you’re tired. And most importantly, I’m focused on you, not on the technology. It becomes just like a partner in that process.
MassDevice.com: Take your example of the Xbox. The technology is there, it’s just not being applied in this fashion. I think there are a lot of instances like that, as you describe.
NVT: So I’ll give you an example, something that really blew my mind. It was simple.
CT images historically just did sections. They are very difficult to process, because we don’t think in terms of slices, so it’s quite specialized as to where you are in the body. The presenter’s lead up was assessing how well his friends could determine where a CT scan slice was taken. Then he finally comes to the punch-line: "We got a iPad and we got an Xbox, it’s a twenty-five dollar item. We integrated it into the CT data and we basically stood them up and allowed them to scan the iPad up and down their body to see the results of the CT scan, contextual with where the iPad was. There was no scanning going on, all it did was say, "Ah, that’s where the iPad is, I need to project that image" as you moved up and down.
To me that was brutally, truly disruptive. It’s not diagnostic, that’s not what it’s about. It’s about interpreting information, contextualizing it.
To be disruptive with voice, artificial intelligence, natural language processing in the healthcare arena, it has got to be contextual.