Clinical notes are one of the most rich sources of data in
the electronic medical record. But using the data stored in those notes is
a very challenging task. There are a number of
different methods and approaches to working
with clinical text, including text mining and
natural language processing or NLP. Although text mining and
NLP are sometimes used interchangeably and they
both analyze text data, they are actually
fairly different techniques. You can think of text mining as
analyzing the raw text data. These methods look to see if
a word is present or absent, or it counts the number of times a particular word
appears in a document. NLP on the other hand is
a more sophisticated approach, and looks at what the words in
the sentence actually mean. It tries to understand more
information about the text and how the words are combined to understand the context about how
each word is used. Let's consider a record
that has the sentences, "The patient does not
have diabetes," and, "The patient's mother has
a long history of diabetes." Text mining approaches might conclude that the patient
has diabetes, because it was mentioned
twice in the note. NLP methods would not call
this patient a diabetic, because it understands
then the first sentence, the word not negates
the word diabetes. In the second sentence
it understands that the person who has diabetes is
actually the patient's mother. NLP is able to understand
these differences because it actually breaks down
the linguistics of the sentence, figuring out the structure of the sentence and how
that structure can effect the context and meaning of
any individual word or phrase. I like to think of
these two methods on a continuum of complexity
and sophistication. Text mining has lower complexity. The methods are fairly straight
forward and easy to apply, but they're not very sophisticated. They miss all the context present in the note and they come
to the wrong conclusion. On the other end of the spectrum, NLP is both incredibly complex
and very sophisticated. It is difficult to
create an apply but has very good performance at understanding the true
meaning of the text. We're going to focus this course
on this middle area. Applying moderately complex and
somewhat sophisticated methods to work with clinical text. These text processing
techniques are useful when you need data that is
only found in clinical texts. You need a relatively
straightforward way to get at that information. We'll apply these methods to two different tasks,
identification and extraction. Identification is where we want to know if something is
true about a patient. Do they have a disease? Did they have a reaction to a drug? Extraction is where we need to pull some piece of information
from the text. What reaction did they have? To accomplish these tasks, we will combine some of the linguistic fundamentals from natural language
processing with some of the techniques of
text mining to create an effective practical approach
to handling clinical texts.