I read an article earlier in the week about the concept of abstracting data from unstructured text as a pivotal concept in healthcare. While it was an interesting read, I was somewhat confused about the use of the word “abstract” in this context.
For those of us who have been English language scholars the natural tendency is to turn to the Oxford English Dictionary with its rich heritage as a source of clarification. Here, the etymology of the word “abstraction” is from the late middle English verb “abstrahere”, meaning to “draw away”, or to arrive at a broader conceptual basis from specific facts.
This is what a physician does through their differential diagnosis: working through sets of hypotheses that are sharpened against the wheel of data until some plausible outcomes emerge that can be categorized as a disease which is in itself an abstract entity.
I’m not sure that this same physician would want a completely automated “abstraction” tool. What they are really looking for is supporting data, in whatever forms it may be found, rather than a broad concept with no context. They are looking for data from general purpose notes that clues them in on the diagnosis, based on the patient’s history and current presentation.
The reality of the world I live in today is that “outside records” are often a messy collection of rich data that require the expertise of clinical staff with advanced tooling. It’s a semi-automated process because the data is dirty, duplicative, and sometimes even contradictory. We’re not looking to pull sets of discrete values, but rather to summarize the patient’s condition in a way that makes sense to the physician based on disease type.
The intent is that a physician saves time by reading the summary, can use the navigation features of the document to easily find supporting evidence when needed, as well as completely avoiding the records they don’t need to see, such as routine blood work, orders, and any other trivia that distracts them from the task at hand.
So, abstracting data from unstructured text may not necessarily be a pivotal concept in healthcare. What is pivotal is that the physician makes the best abstraction, i.e. diagnosis, based on having timely access to the complete data set needed to make that decision.