Natural language processing (NLP) is an important aspect of artificial intelligence that allows machines to process large amounts of data in human (natural) language – for instance, open-text fields or speech-to-text data. With NLP, just as with human language, it is critical that the machine is able to distinguish three things: content, concept and context.
Content: the substance or material dealt with in a speech, literary work, etc., as distinct from its form or style.
Identifying content is merely having the NLP pick up certain words or phrases.
Concept: an abstract ideal; a general notion.
Context: the circumstances that form the setting for an event, statement, or idea, and in terms of which it can be fully understood and assessed.
Concept and context are where artificial intelligence really shows its power. The machine needs to have the ability to distinguish the word or phrase as it relates to the content as a whole, and potentially compare that with other data (such as historical notes) to make a determination.
A great example of this is the abbreviation or acronym, “ID”. ID could stand for a myriad of concepts depending on the context in which it is used, for example:
- Infectious Disease
- Infectious Dose
- Iron Deficiency
Hospital-specific abbreviations and regional dialects in a patient’s record can also vastly skew the original intent of a caretaker’s note.
As the “patient narrative” is mainly unstructured free text it is often difficult for the care team to sift through the volume of historical data available on even one patient. Artificial intelligence and NLP tools can be used to quickly scan through massive amounts of patient data and identify relevant textual components to surface that may be relevant to a diagnosis now. But how does it know what the doctor or nurse was actually trying to say?
“There is a ton of unlabeled data in the electronic health record, says Michael Seeber, Data Scientist at Pieces Technologies, “what our team does is work to make clinical sense out of it.”
Improving AI models, or, fine tuning them to the nomenclature of your health system and staff is where human-assisted machine learning comes in. At Pieces, our employed physicians and data science team work together to monitor the NLP with a goal of operating at or above 90% sensitivity and positive predictive value (PPV).
So how does a clinician train something as complex as an AI model without knowing how to write and decipher code? Enter the Pieces DS Learning Lab.
The Pieces Technologies DS Learning Lab is a studio that enables Pieces-employed physicians or your hospital’s clinicians to enter terms or phrases to understand how the Pieces AI is deciphering its importance. Terms that are more important, or applicable, for a certain hospital can then be used to help train the model. The Learning Lab also allows the Pieces data science team to leverage learnings across hospital environments much easier and faster.
As hospital needs change over time and your EMR adds new patients and data, you need a decision support system that can keep up with your ever-evolving environment. The DS Learning Lab is just one way that we are making technology work better for you.
This two-part blog series will introduce you to our Learning Lab and show you how we:
- Can use Artificial Intelligence and Natural Language Processing to mine old data for new use cases, and
- Quickly transfer these learnings across health systems, and how we adapt quickly to institutional change