Kyle Silvestro is founder and CEO of SyTrue of Chico, CA.
Tell me about yourself and the company.
I’ve been in the world of clinical natural language processing and semantic interoperability for the last decade. My team collectively has been in the industry for more than 45 years.
As a company, we focus on the world of data. We look at ourselves as an oil refiner, taking all the data that’s being created — transcription, dictation, typed notes, structured order entry, what have you — and creating a refinery process that we put it through. On the other side of that, we get structured data that’s semantically interoperable.
We focus on that pipeline that allows organizations to create normalized data to drive down to processes like analytics, decision support and population health.
People often get natural language processing confused with speech recognition. Describe NLP.
It’s the ability for the computer to go through a written document — a Word document, PDF, or something that is the by-product of speech recognition – and recognize and understand the content. Not only the content, the meaning behind the content as far as it’s something positive, something negative, or something concerning. Beyond that, be able to make decisions as far as how that should be encoded with a terminology or medical knowledge base such as SNOMED, ICD-9, or ICD-10.
I’m a huge fan of keeping the clinical narrative and patient narrative and not just discrete data element factoids. Is there a demand for that?
It’s interesting what’s occurred over the last decade and really the last several years. Data has become important and incentives are changing to where they’re making data much more relevant in the chain of care. As organizations are looking at this, they’re looking at a lot of claims data, which gives you an incomplete picture.
Until you start marrying the clinical narrative with the claims data, you are not going to see the outcomes or the population that needs to be managed comprehensively as you would just looking at a single point of data. The market is realizing that the data is important and the data is the key for them to being successful.
How good is NLP’s inference capability in reliably turning free text into discrete data?
That’s a question we get asked frequently. My response back is, how accurate is the physician’s note? At times, and depending on where you are across the nation, the note may mean different things. Words may mean different things, context may be a little bit different.
It’s about being able to create a ability to normalize that information and then continuously learn on top of it. Create a feedback loop of this data to ensure that the inferencing or accuracy gets extremely high. Once it’s extremely high, you can build some rules around that to flag inconsistent actions or items that may not be just exactly right for manual review.
It’s great for a number of different processes, but there are still some situations like Core Measures or others that do require clinical opinion. In that context, it assists organization significantly and it’s highly accurate.
Google Flu Trends stopped working because it was measuring indirectly captured data that Google didn’t control or understand as it changed. Is that a risk in using NLP to analyze EMR data of a somewhat uncontrolled origin?
No. You have to put it through a process where you can turn data into semantically interoperable content, to create a process that fits an organization and its work flow.
I’ve been at one hospital and seen 152 different ways that they document the section heading of medications. In one hospital. How do you give organizations the ability to normalize that data and to ensure that the section heading of medications corresponds to the appropriate LOINC code and that all these 152 ways all roll up to a single code of medications, if that’s what the organization desires?
It’s about giving them the ability the look inside a black box that was formerly called NLP and terminologies and being able to use that information in line with the organization’s objectives, work flows, and outcomes. Each document can have a different purpose in life and have a different recipient in life based upon on the data that’s within it. Being able to give organizations that flexibility that they haven’t had in the past to be able to perform actions like this changes the paradigm and maybe the questions that are being asked.
What can end up is organizations get to highly accurate data that’s interoperable, that drives downstream processes, can identify patients that are at risk for medication non-compliance, and a whole other host of activities that are either going to reduce cost, help alleviate risk, or identify opportunities for revenue.
You mentioned that the system can learn. How does that work?
In the case of ICD-10 right now, it’s a documentation issue. A lot of the problems that we’re facing in healthcare come back down to documentation. It may not be as sexy as some of the other topics that are out there, but at the end of the day, if you can get to the point of care with a document or parts of documentation are being created, what you’re doing is able to add almost real-time support into that encounter, or creating something along the lines of a encounter-based analytics. As you’re moving forward in this process, it’s about identifying the points in the work flow that can make a difference to have that impact that you’re looking for.
I think the answer really is yes to your question. Organizations are seeing that value.
How much setup is required to get the information that you need from the EMR and to figure out its structure?
The US government is, I think for the first time, focusing on standards. If the laws around Meaningful Use are still upheld in October, that standard’s going to be Direct over the Blue Button. If you’re able to then able to pull information out of these standards, process it, put it together in a consolidated CDA, you’re able then to hand that off to the next person in the chain.
If organizations start complying with this thought of interoperability and data mobility, we all — vendors or third parties to the record or to the process – can help move forward this continuous care to increase outcomes and value within the healthcare system. Their thinking, and what we find, is the closer we go to the data, the easier it is, and the further away, the harder it becomes. We end up pretty close to the data source.
Going forward, we’re anticipating this model where we can get that in real time via a standards-based approach that would allow organizations to create something like a meta layer or meta data of smart intelligence. Then the EMR and HIE that can add value into that record in real time.
Organizations that work with us are up and running within an hour more often than not, minus some of the interfaces that they have to create.
What are some examples of what people are doing with your system?
Organizations are looking to identify populations that may be at risk for heart attack or stroke. They are looking through their more often than not transcribed documents, because these are high-value specialties that use maybe a limited piece of an EMR to identify patients that might have been missed or have not been recalled in a certain period of time to follow up for a visit.
We’re being used to look at site selection for clinical trials, by being able to identify possible patients that would fit within a certain selection. Other areas to alleviate risk, or feed data into third-party systems to assist their predictive analytics, decision support, or business intelligence. We act as a platform across different organizations so they can send data and have it refined, processed, and get that refinement back out in order to add value to what they’re currently doing.
You compete with least one or two big companies that offer NLP-based services, including Nuance and its Clinical Language Understanding. Why is your product a better choice?
There’s a very large untapped market. It’s a matter of focus. We’re heavily focused in areas that Nuance isn’t and we’re able to add value along those lines.
As I look at the industry and I look at the last 10 years of being in the business, I’ve probably failed more than most in failure of sales, but I’ve also been quite successful. I think I’ve come to understand the bottlenecks and the impediments and the push-backs that have always been around clinical natural language processing. I think we’ve addressed those and we’ve focused on those points.
Building that into our pipeline and workflow that will allow both a rapid adoption and a platform-type view of this data, where many people can tap into via a Web service-based approach. It will utilize technology that gives them the ability to do natural language queries and then to be able to bring a refined data set into any one of their processes.
While there’s a lot of competitors out there the market and a lot of new companies emerging, I think it’s the collective 45 years of experience my team has that give us an advantage in the way that we look at the marketplace and the solution that we’ve brought to bear.
Where do you see the company going forward?
We just started releasing the product commercially. We’ve been hand-selecting our clients and beta sites to ensure that we have something that is meaningful that will make a difference in the market.
When people looked at it, they’d say wow, I’ve never seen anything like that. HIMSS was the first time that we started showing that off. That’s kind of the response that we’ve gotten, at HIMSS and almost every other discussion that we’ve had.
The company is focused on methodically growing its client base and delivering beyond expectations to our current users. We’ll continue to add clients based on our reputation and our delivery.
Do you have any final thoughts?
We have a very interesting time in front of us. The world and specifically healthcare is opening up to the idea that clinical documentation is important. It’s the needle in the haystack. If you can look there, you’re able to look across the longitudinal record and add value to the people’s lives who matter, who feel like the forgotten soldiers in this, which are physicians and patients. If you can remove the impediments and barriers to that, everything will go forward and healthcare will be a fundamentally different place.