Natural language processing can improve clinical documentation, drive actionable insights and help reduce clinician burnout, one expert says. But providers must be on board.
Natural language processing, or NLP, is a branch of artificial intelligence that enables machine understanding of human language. In recent years, NLP has become part of consumer products and performs with a sometimes surprising degree of accuracy.
Virtual assistants use NLP to understand spoken questions like, "What's the weather like today?" They can contextualize the meaning of the question, and then use data from online sources to reply with a meaningful response.
Medical NLP has long been a topic of research and development, since it contains significant potential for uncovering meaningful insights from unstructured medical text. This presents huge opportunities for healthcare providers, clinical researchers and payers to discover valuable information related to disease progression, treatment efficacy, population health trends and many other use cases that would have been infeasible to identify by using manual data-review and analysis techniques.
However, development progress has historically been slow due to the complexities inherent in medical language.
Healthcare IT News sat down with Dr. Tim O'Connell, a practicing radiologist and CEO of emtelligent, a medical NLP technology company, to talk about how NLP technology works with human prose, how NLP still is poorly understood, and how medical NLP can only succeed if it caters to clinicians.
- Please break down NLP for our readers.How does NLP technology read, understand and structure human prose (for example, physician documentation)? Further, what does NLP then enable clinicians, technologists and researchers to do?
- Natural language processing has existed since the earliest days of computers. NLP combines computer science, linguistics and artificial intelligence to create software that can understand or even create human language. Today we'll be talking only about NLP software that can understand human language, not the kind used to generate text.
NLP software uses multiple methods to read text and "understand" some or all of the content it is given. A good example in the medical field is why searching electronic health records without NLP can be very difficult.
Let's say a physician wants to search a patient's chart to determine whether they have hepatitis. Using only a keyword-based search might return tens or hundreds of hits that aren't relevant, such as a family history of hepatitis or whether the patient had been tested. But if clinicians could use NLP software to read and understand a patient's chart, they could search much more effectively. It would be akin to asking a smart assistant, "Does this patient have hepatitis?"
NLP solves the scale problem of needing humans to read medical text to understand it. When you can use computers to understand large amounts of medical text, you open up new use cases for things like automated adjudication in insurance, predictive analytics, closing gaps in care, developing new AI models, better clinical applications for care providers and more.
- NLP is groundbreaking in healthcare. But you suggest the technology still is poorly understood, with confusion on how NLP and AI work together. Please explain.
- Yes, there still is a lot of confusion regarding NLP and AI. While NLP is a field under the broad umbrella of artificial intelligence, when people think about capital-A capital-I "Artificial Intelligence," they too often are thinking about computers that: 1) Independently make decisions to perform classification or prediction tasks, or 2) Interact naturally with humans.
NLP typically is not just those things; instead, it's a required component of nearly any AI "system."
For example, to use the most "Hollywood" of use cases, when a robot has a conversation with a person, NLP is used both to generate the robot's speech and to understand the person's responses. Applying those capabilities to our use cases, if your goal is to create a medical AI model that predicts recurrence of a patient's cancer, you would want to use NLP software to curate good training data, so the model was getting the information it needed.
If you're training a radiology model to find pneumonia on chest X-rays, you might need hundreds of thousands of images labeled with pneumonia, so you would run NLP software on radiology reports to find the cases positive for pneumonia to get your training set.
In short, NLP is both a type of AI and a tool used "behind the scenes" to create and improve AI software. NLP software will be used to extract training data from patient charts to create new AI models that can predict and prevent medical errors and disease, improving outcomes for patients.
- You contend that NLP software can only succeed if it caters to clinicians, something not yet truly achieved. Please explain what you mean, and what it will take for NLP to get to where you think it needs to be.
- There is a saying that "medical school is a four-year terminology course," and it's true. I mention this because I really believe we need good medical NLP software, created by clinicians and NLP experts working hand in hand, that truly understands the content of medical text.
One issue I hear about from clinicians is that older NLP systems don't read the text like a doctor or nurse would; they say it misunderstands words, doesn't pick up on negation properly, or is confused by lists in reports.
When people use truly great NLP software that can understand the original meaning of medical text, a whole new world of possibilities for improving our health systems and patient care will become available. NLP can be used to create new applications such as automated patient summaries, as well as smart search and documentation tools that enable them to spend more time with patients and less time sitting in front of screens.
By applying NLP to data science and analytics, healthcare facilities, payers and governments will be able to get higher-quality data about patients. Some of our healthcare system inefficiencies are due to lack of data because it's too expensive to pay people to extract it from charts.
NLP solves this problem and can give administrators and analysts the data they need to make better, more informed decisions, more quickly.