Growing numbers of healthcare executives recognize the broad and critical uses of Natural Language Processing (NLP). To learn more, we sat down with SyTrue Founder and CEO Kyle Silvestro to discuss NLP’s key role in healthcare now, and its future direction.
Big questions about Natural Language Processing (NLP) are fading away but a few remain. How do you define NLP’s capabilities when someone asks, “What is NLP, exactly?”
NLP allows computers to read, understand and extract in-context information from free text (like Microsoft Word or written notes). With NLP, we can normalize and extract data and turn it into interoperable, actionable information – which helps patients and providers equally.
Bear in mind that this is a mature technology – the result of highly accelerated HIT innovation. NLP in particular has vastly advanced in the past decade. Comprehensive NLP systems, in fact, show how far HIT has come.
The sophistication of NLP is now clearer to many in healthcare. We experience NLP when we use Siri with a phone. But take us back to NLP’s early days. What did the first NLP applications in healthcare do and how has that evolved?
NLP has evolved significantly since the early days, both in solution design and in functionality. When NLP started, it was – often still is — a closed system. NLP engines didn’t have smart user design or good interfaces. They came in programming languages like Prolog, usually from universities. These were quickly outdated and hard to build on as scalable platforms. Most NLP engines worked well with the training sets used to “train” or create the technology, but those engines were never designed as scalable solutions, which is why NLP has been difficult to adopt.
The first technologies were not architected to handle hundreds or thousands of physicians from different specialties, all with different speech patterns, abbreviations, misspellings and the multitude of different document types.
Successful solutions focused on narrow uses like computer-assisted coding (CAC) in one clinical area (like radiology). Because of that narrow focus, those solutions were unusable beyond that context and were hard if not impossible to adapt in other clinical domains. What’s more, CAC products are not good general clinical coding solutions since revenue cycle management requires only a portion of relevant clinical information to create a billing code (the remaining clinical information stays un-coded).
But today’s needs grow ever more complex. Any NLP engine must be able to identify every piece of data from every document type – a big demand on any simply-designed NLP technology.
NLP has sparked the view that it’s a primary way to manage the waves of new unstructured data (texts, notes and more). NLP shows real promise in reducing errors, cutting redundancies and driving evidence-based decision making, all crucial for healthcare. What will make NLP broadly useful in healthcare?
Though we’re experiencing accelerated innovation, some NLP “solutions” tend to be partial ones. If you have just an NLP engine, chances are you won’t succeed. You need more than an engine, just as a car does to run properly. It’s the same with an effective NLP solution.
It takes a full system, starting with a workflow solution offering a broad range of content when a caregiver or data scientist needs it. You’ll need something like our SyTrue NLP OS (Operating System) ™, a platform that accounts for variations in language, spelling, coding requirements and document types. You also need an integrated, scalable platform to assess accuracy of information, because manually handling the thousands of documents created daily is costly and a time killer. This includes a semantic rules engine coupled with an advanced terminology server to be administrated by end users – so NLP can support care delivery. With our NLP OS ™, we enable our partners to create their own NLP-based rules. They don’t depend on SyTrue for updates. So if a billing company uses a computer-aided coding (CAC) product and wants new rules, it creates its own coding rules to automate the process. This is the power of a comprehensive solution like SyTrue NLP OS ™.
Will smart use of unstructured data move us beyond healthcare’s structured data platforms? Or will NLP work alongside those data platforms?
There’s another way to think about this. It’s time to move away from unintelligent and poorly designed structured systems that lower productivity and add little value to patient care. By letting providers document in free text – as they’ve always done — you preserve the clinical narrative and with NLP, you create actionable data at the point of documentation. This data is usable for managing many groups of patients, and NLP also reduces the high error rates in health records that appear in non-NLP solutions.