Consensus Cloud Solutions has added Natural Language Processing capability to its offerings which will make it much easier for healthcare organizations to turn unstructured documents into a rich source of information. Their added capability means that clinicians will now have additional context when providing care to patients, which can lead to better outcomes and reduced costs.
NLP + OCR
Optical Character Recognition (OCR) is the technology that turns handwritten or typed text into machine-encoded text that a computer understands. OCR is typically applied to scanned paper documents, photos, and other electronic images.
A limitation of OCR, however, is that it’s not capable of providing context for the documents it processes. Here is an analogy: Think of OCR as an Italian translator who is completely unfamiliar with baseball who is asked to translate a baseball broadcast. Although they might be able to convert the words being said, their lack of context would result in a translation that might not be completely understandable. How would you translate a “double-play” for example if you didn’t know what that actually was?
Natural Language Processing (NLP), allows computers to “understand” the contents of documents by analyzing the words and language used. When applied successfully, it can extract information and insights from those documents.
OCR + NLP is a powerful combination when applied to electronic documents, including faxes which are still used frequently in healthcare.
Consensus Cloud Solutions’ Clarity offering combines OCR and NLP in a powerful tool that unlocks the unstructured data that is held within faxes.
Extracting Value
“When a fax is rendered at the receiver’s end it is difficult for that information to be put into a database,” explained John Nebergall, Chief Operating Officer at Consensus Cloud Solutions, in an in-person interview with Healthcare IT Today. “What we attempt to do, using Clarity, is structure that unstructured document and bring it into the [medical record] in a way that’s meaningful. and allow it to be used for a better patient experience.”
The goal is not to simply translate everything that is on the faxed document, but rather understand what that document is, and extract the meaningful information that a clinician can use.
Practice Makes Perfect
In order for an NLP solution to be effective, it needs to be trained on a sample set of document. The bigger and more representative the sample is, the better the solution becomes. Consensus processes millions of faxes for healthcare organizations through its cloud-based fax solution which means their NLP engine has plenty of practice. Even better, as Consensus customers find new uses for Clarity (ie: new problems that it can help solve), the system can be quickly retrained on this massive dataset.
“What we’ve learned is that in order for natural language processing to be truly effective, it has to have a lot of practice opportunities to be able to learn from it,” said Nebergall. “We have literally billions of pages of fax that flow across our network on a regular basis, giving us an opportunity to train the system billions of times. That’s our advantage.”
Watch the full interview with John Nebergall to learn:
- What the manual process of interpreting a fax is really like
- Why NLP is being widely adopted now vs 3 years ago
- Where unstructured documents may be hiding in your healthcare organization
Learn more about Consensus Cloud Solutions: https://www.consensus.com/clarity/
Listen and subscribe to the Healthcare IT Today Interviews Podcast to hear all the latest insights from experts in healthcare IT.
And for an exclusive look at our top stories, subscribe to our newsletter.
Tell us what you think. Contact us here or on Twitter at @hcitoday. And if you’re interested in advertising with us, check out our various advertising packages and request our Media Kit.
Consensus Cloud Solutions is a proud sponsor of Healthcare Scene.
Transcript
Transcript
[00:00:08] Colin Hung: HI, I’m Colin Hung with Healthcare IT Today. And I’m excited to sit down again with John Nebergall, Chief Operating Officer at Consensus Cloud Solutions. John, welcome back to the program.
[00:00:27] John Nebergall: Colin is great to be here. Thanks for having me.
[00:00:29] Colin Hung: So today we’re going to be talking about unstructured data and how it’s an untapped and undervalued resource. And we’re going to talk about how you can unlock it.
[00:00:39] John Nebergall: Great. Unstructured data is my favorite.
[00:00:43] Colin Hung: I know you’re just saying that for this interview, but I love it. I love it. So let’s start with a little bit of context. What is unstructured data and why is there so much of it in healthcare?
[00:00:52] John Nebergall: Well, if you think about the way modern databases work – structured data is data that loads into a database, in certain fields so that you can search that database, pull data out in certain contexts and are able to use analysis to understand what that data is telling you.
[00:01:11] Unstructured data is data that doesn’t fit that mold.
[00:01:15] So if you think of words on a piece of paper, for example, that’s unstructured data and a database can’t really use it. If you think of things like fax…when a fax is received by a healthcare organization, that’s loaded with unstructured data – can’t be used. Doctors handwritten notes, unstructured data. So that’s really the problem in healthcare: how do I get to the goodness of that unstructured data and put it in a way that can be searchable, analyzed, and can help us with patient care.
[00:01:45] Colin Hung: Yeah. I was going to ask about that. So how can this unstructured data actually be used? I get how you can read it and interpret it, but in terms of turning it into something that computer can use, what are some of those use cases for the unstructured data?
[00:01:58] John Nebergall: You know what, you’re hitting the nail right on the head! As human beings, we can read that document and in our head, we structure that data. We can understand it. We know what it means. Computers can’t do quite the same thing. I’m sure that you’ve seen from time to time a poor data input person with a stack of paper typing stuff into the computer.
[00:02:20] It’s exactly that kind of manual process that needs to occur. If you think of the days before Consensus, to take that unstructured data, structure it and make it useful. What we essentially do is apply technology called machine learning, artificial intelligence, to understand almost like a human does, what the words on that page mean. By doing that, we can extract the important data, structure it to be loaded into a database, and actually make it useful and searchable.
[00:02:54] Colin Hung: So what you’re talking about is natural language processing, NLP. Is that right?
[00:02:57] John Nebergall: That’s exactly right.
[00:02:58] Colin Hung: And so that is able to read the document and extract the relevant information. So it’s not turning the whole document into an electronic format, right? It’s sort of interpreting the important stuff that you want to get out of like the handwritten note or the fax or those kinds of things.
[00:03:12] John Nebergall: That’s exactly right. The database is looking for certain things. As a provider, you want certain pieces of information. You don’t necessarily want the whole thing all at once. You want what you’re looking for. Natural language processing works the same way. You say “I want to extract this kind of information” as you understand the document from the automated intelligence…that intelligence extracts that information and puts it into the database so it can be referenced later.
[00:03:40] Colin Hung: Now to me, that was something that came out of the session that your company was just doing. That was something that struck me – you don’t have to convert everything off the document. Your goal is not to digitize the entire document. Your goal is to interpret it, to really grab the information that you want and need…and leave the rest.
[00:03:57] To me that was a unique perspective. I never really thought about that. When I think of NLP, in the past, I thought, you want to interpret the entire document. But what I learned was that’s not the case. You’re actually looking for something very specific and therefore it actually makes it more accurate.
[00:04:11] John Nebergall: Right. Data is data. Relevant data is information. That’s really the goal here to get the information.
[00:04:18] Colin Hung: Now, you recently announced the debut of your NLP product. Do you want to tell us a little bit more about that?
[00:04:26] John Nebergall: Certainly. Consensus Clarity is the product that we’ve created specifically with fax in mind. We understand that so much information is transferred inside of healthcare using fax.
[00:04:40] But when that fax is rendered at the receiver’s end, very often, it’s difficult for that information to be put into a database easily. Sometimes it’s rendered as a piece of paper – it goes through that process of getting manually entered. Sometimes it’s just attached to a patient record as a document, but not easily searchable.
[00:04:58] So what we attempt to do, using Clarity, is structure that unstructured document – bring it into the database in a way that’s meaningful and allow it to be used to improve the patient experience.
[00:05:11] Colin Hung: So if I was a CIO or a CMIO or any health care leader for that matter. What are the signs or triggers that I might look for to know that I have some unstructured data, either a challenge or an opportunity to tap into unstructured data that I haven’t before?
[00:05:30] John Nebergall: Filing cabinets is a good cue. If you see that you have fax machines in various parts of your organization, that’s going to be a cue. Anytime you have paper stacked up, that’s a cue. All of these things, and I’m sure if you walk through a healthcare organization it’s pretty easy to quickly see these kinds of cues and say – look, there’s a lot of data out here that is useful, it’s just not being put in a context that can be used in the patient encounter.
[00:05:59] Colin Hung: I’m assuming as well, a lot of printouts. If you see a lot of papers just on the desks and things, that’s another sign that I’ve got a lot of unstructured data here.
[00:06:09] John Nebergall: Yeah. That’s exactly right. And all of that unstructured data is potential, right. It’s how we can close the gap between where we are now and a fully informed physician who’s treating a patient at any given time.
[00:06:22] Colin Hung: It just surprises me that we have so much information locked in this unusable format, this unstructured format. And what you’re talking about is really poignant because yes, we still have fax machines, we still have people printing things out, we still have paper documents. And what you’re talking about is the ability to take that in and make it useful for the clinician or the patients for that matter.
[00:06:42] John Nebergall: Exactly. And when you do that, you really start to unlock some of the things that are in the shadows that can be helpful in being able to treat a patient and right now just don’t have that status.
[00:06:55] Colin Hung: Now, NLP is AI. It’s under the umbrella of artificial intelligence and AI. And I think a couple of years ago, we reached the top of the hype cycle. AI was being promised that it could do lots and lots of things. I think we’ve come down off that peak because we’ve had some high profile failures of AI. Where do you think AI is now? Are we at a point where AI is a bit more realistically positioned in healthcare? And if we got some more realistic use cases for it?
[00:07:23] John Nebergall: Well, I think two things have happened. Number one, technology has advanced and number two, we understand better how to be able to apply that technology to the problem.
[00:07:33] So I think what we’ve learned is that in order for natural language processing to be truly effective, it has to have a lot of practice opportunities to be able to learn from it. That’s where that machine learning comes in. And the more that you’re able to present the machine, the more you’re able to show it how you correct it as a human being, the machine remembers that.
[00:07:53] So it starts to be able to do what a human does. The key is you have to be able to do it over and over and over again. And that’s really one of the advantages that we have. We have literally billions of pages of fax that flow across our network on a regular basis, giving us an opportunity to train the system billions of times. That’s really where our real advantage in this comes.
[00:08:21] Colin Hung: It sounds like it’s just more “proven AI”. It very robust and we know it works. And like you said, you’ve been able to do this thousands and thousands of times in a repeatable, predictable manner. For AI, that’s what you want to get to.
[00:08:33] John Nebergall: Absolutely. Absolutely.
[00:08:35] Colin Hung: So John, where can people go to find out more information about Consensus?
[00:08:39] John Nebergall: You can visit us at consensus.com and get all the information you need there,
[00:08:44] Colin Hung: John I really appreciate all the great information as always. It’s a joy to have you on the program.
[00:08:48] John Nebergall: Thank you so much for having me. I really appreciate it.
