The Fundamental Fracas in EHRs: Narrative vs. Structured Data
“We were probably mistaken to think of words on screens as substitutes for words on paper. They seem to be different things suited to different kinds of reading and providing different sorts of aesthetic and intellectual experiences. Some readers may continue to prefer print, others may develop a particular taste for the digital, and still others may happily switch back and forth between the two." -Nicholas Carr Paper versus Pixel
We live in a time when electronic representation on-screen is the chief means of recording and communicating experience. Whether still, video, or text, our ever-increasing interaction with digital screens carries with it the uninvited companions of time constraint, peripheral bombardment, and reduced ability to linger and consider meaning.
For all of human history, the effort to record is intended to put forth story. Long ago, narratives were written on memory itself, and before the creation of writing--perhaps 5-7 thousand years ago—they were transmitted from one individual to another orally. However, with writing came a means through which records could be kept on objects such as stone, metal, wood, papyrus, and, of course, on paper. The symbols and languages in which these records were kept evolved in various streams and patterns into the written languages of today.
Now we have a new medium of storage, the database. It functions, for all intents and purposes, not unlike memory in the ancient oral tradition, as it has the capacity to store information for recall. There are key differences; however, between information stored in human memory, and information stored as electronic data. Chief among these is highlighted by memory’s ability to parse, synthesize, and paint image out of data; it can embellish, alter, inflect, musicalize, and otherwise impart to the information characteristics that are absent in digital information. There, data can be extracted, but it does not have the human quality of story. Attempts are made via software to give it such a quality, but these always fall significantly short. Of course, story can be stored as narrative itself, in the form of, say, voice bytes, readings, movies, text documents, and other forms in which a human has made the effort to create synthesized data forms that impart intended (and unintended) images to the viewer or reader. Hence, a single datum may be a narrative.
However, once a narrative has been broken down, and data has been extracted from it, that data cannot be brought back into an accurate narrative, except by human intervention. This is not unlike Humpty Dumpty in the English nursery rhyme:
Indeed, the abstraction of any narrative into points of data is a great fall.
And if Humpty represents the meaning inside any story, once extracted, it is not possible to put back together again...
Humpty appears in Lewis Carroll's Through the Looking-Glass (1872), where he discusses semantics and pragmatics with Alice.
Alice was too much puzzled to say anything, so after a minute Humpty Dumpty began again. "They've a temper, some of them—particularly verbs, they're the proudest—adjectives you can do anything with, but not verbs—however, I can manage the whole lot! Impenetrability! That's what I say!"
...one datum plus another datum plus another, and so on, may not really be brought together into a true narrative, except by human intervention.
It is the telling of, and listening to, story that remains embedded in the human psyche as the common ground sought by both recorder and reader/viewer, and through which one human conveys (x) to another. The axiom “a picture is worth a thousand words” is quite familiar; adding “and a story includes ten thousand pictures” illustrates the huge chasm that electronic records face, between story and elemental structured data.
Physician par excellence and renowned philosopher of medical practice, William Osler, often reiterated to his students and colleagues “listen to the patient.” In many ways Osler’s was a plea, an appeal with evident foresight lamenting a future time when information residing with the patient would become less important than that in the chart. That time, the post-Oslerian era, has arrived. Indeed, listening is so central to medical care that failure or inability to listen often lays at the feet of many medical liability cases. It also sits at the center of patients’ complaints about a healthcare system that seems to lack heart. “My doctor just won’t listen to me.” “I told him many times.” How often have we all heard these petitions? Is there some underpinning, some systemic cause, some embedded characteristic of modern medicine that explains why this should be the case? And can anything be done about it? Should anything be done about it?
Let’s examine the premise that structured data has become too preeminent in the medical realm, has established hegemony shoving story aside. The process of this revolution antedates electronic records, and began overtaking the conceptual, cognitive, framework of physician-patient relationships as quantitative laboratory and definitive imaging information rose to occupy the minds of providers. And yet, demise of story, demotion from its justified first tier in open communication, naturally cripples listening ability.
Think about it.
Are medical providers alone in being crippled this way? It seems not. Indeed, deafness to the “real story” behind any human encounter is not isolated to medicine or medical records, but has become an embedded, pervasive, and characteristic of our technological society. Even social media screens off one individual from another.
The impact of deafness pervading our lives is great: for when another does not hear, it feels to almost everyone within range that caring is thoroughly absent. Hardly anyone is going to tell you that our society is a caring one. Hardly. Not only medical providers, but also the service people with which we transact life, are seemingly deaf. From cashiers to accountants, from clerks to lawyers, all are unable to hear about us, about why we have approached them—at this time, at this moment-- and about what really matters to us. Their inability to hear does not mean they are unwilling to hear. Just, perhaps, possessing insufficient tools with which to make the connection.
Instead of the connection of story, our lives are given representation in the new psyche of modern tech as merely data points; our narratives are lost, unspoken, or disparaged as lower class inhabitants of a data-rich informational world. What, otherwise, is the source of our hesitation as we approach strangers with whom we need to interact? With each desk, each clerk, each individual that we’d rather not encounter arises an unwanted dread. Will we be “dissed” as if lower class due to whatever need we may have? Will we bother the soul, the concentration (the texting?), of another when they are, due to our presence, required to extract from us data?
In fact one may often find that a reason medical providers are sought is because it is one place where hope exists to find a listening ear.
Whether that hope is realized or not is an unanswered question in every case. And yet how many of you have witnessed the disappointment of those seeking care, only to be given the perspective that the provider did not really care? So one wonders: was something more present in the disappointed physician-patient encounter than just bare-bones search for data on the provider’s part? Or were the underlying forces perhaps hidden. Acquisition of structured data in the name of caring, while peering into-but not hearing--the story, has the potential to hide the truer hearts and minds caring under a cover of seeking to acquire data in its name.
Now, even a layperson knows that anyone presenting for medical care—whatever the setting--has a story. Actually, they always have multiple stories. And most often we medical types focus on two main themes: the story of events leading up to the Emergency Department (ED) or clinic visit, and the relevant back-story, or history of key past medical issues.
Let’s take an example. A patient presents to the emergency department icteric and febrile. On the Electronic Health Record (EHR) or perhaps even on a paper template, and elevated temperature is recorded, moreover the triage nurse notes on the chart that the patient appears yellow. The nurse completes a triage checklist, and notes therein that the patient has no prior history of liver disease, but drinks some. A hepatic etiology is first in the nurse’s mind. What then?
In this example, (somewhat exaggerated-but taken from an actual case), the patient is sent back to the main department (after several hours wait of course) and a physician assistant sees the patient, takes a history, and performs an exam. The same data elements (icterus, fever) are discovered, and lab is ordered.
Now the physician’s assistant did a quite good job of taking a history, and in fact the patient told the PA that he had just gotten off a flight from Nigeria, where he had been serving with the Peace Corps.
There was no box on the template “just got off a plane from Nigeria” nor one for “recent travel”. But a rather full template based on the chief complaint of “febrile illness” was nevertheless completed. Moreover the PA included in a section of the chart a written note that included the patient’s recent travel.
The very excellent clinical decision support system the ED had purchased, reported back that a workup for viral hepatitis, biliary tract disease, and other disorders that elevate the bilirubin, should be done.
In the meantime, the patient’s lab work returned, his temperature normalized, and (having no insurance) was sent to a gastroenterologist’s office at the beginning of the following week...
About six days later the ED gets a call from the GI office complaining that they were not told about the patient’s international travel and exposure, and so were impaired in making a diagnosis, and initiating treatment for what was clearly a case of malaria. The patient’s condition had become much worse, and he had been admitted to another hospital as an emergency.
What happened? Clearly this is a case where the medical record and the related medical decision, judgment, and approach were adversely colored by the gulf between narrative and structured data. Data was recorded and attended to, but the fundamentals of the patient’s story, even though recorded in narrative form, was missed.
Attempts have been made—and they abound--to extract data from narrative. And similarly, considerable effort has been directed toward trying to construct a narrative out of the data. Both exist, both have failed. When one reads a narrative built from data, it is, of course, not unlike trying to understand the phraseology of a robot. It is rather unclear, rather disjointed, and is absent in many key details. On the other hand, attempts to accurately extract data from a narrative are recipes for disaster. A case encountered recently is one in which the physician dictated via voice recognition software that the patient was complaining of “low chest pain.” The EHR in question extracted a data point like this:
CHEST PAIN? [ ] [ X ]
No chest pain indeed. The lawyers loved this one, especially after the patient died of an MI after an inferior wall MI.
How can this issue be addressed?
The only hope to bridge this “great divide” between narrative and data is to face it head-on, embrace it in a sophisticated manner, to do so with tools that are appropriate for the task, and then analyze and evaluate the state of our capacity in a realistic fashion. Do any such tools exist? Perhaps. But where?
The simplest tool, and one that has been successfully employed by some paper templates, and EHR spin-offs, is to create and retain BOTH narrative AND data. This functionality has been called a “harmonized” EHR. Terminologically, this is to be distinguished from the use of “hybrid” EHRs which (like hybrid cars) indicate systems in which a provider can have one foot in the electronic world, and the other in the paper world. Granted, when a provider is not experienced in this methodology of using a harmonized system, it may feel a bit cumbersome. However, when narrative thought of as a non-structured data (like a picture or other image), and retained, without trying to extract contained embedded elements, an effective “work-around” results. Such an approach is not without its own problems, however. What do you do when the narrative and data are both present, but simply do not harmonize? What if the narrative from the above example says “the patient was complaining of yellow skin.” And the data says:
ROS- Dermatologic system:
Skin Yellow? [ ] [ X ]
Such problems are common in systems that try to carelessly harmonize data and narrative, and they are fertile ground for plaintiff’s attorneys. And yet harmonization permits the creation of a solid chart if the provider takes the time to be sure the narrative and data correspond, and if the EHR allows this to happen easily.
So where can we extend our technology to do better in this arena?
A good start may be to think of the issue as one not unlike language translation. Say, Swahili to English. Or Greek to Afrikaans. Language translation is certainly a tricky business, whether between spoken and written languages from disparate lands, or between the mental structures by which we communicate (image, narrative, data).
Can formalized principles of translation, give us usable criteria to bring these two different medical languages, narrative and data, into closer harmony? Here is what would, at minimum, be necessary:
Accuracy. If structured data is extracted from a narrative text, the meaning should be accurately and without error reflected in the data that is extracted. On the other hand, if a narrative is constructed in the other direction, from data, it should be represented in a fashion that does not unnecessarily embellish the story with items or words that could result in misinterpretation. That means that, for example, all or many key articles of the English language would need to be left out: no, yes, and an, the, etc.,…..And yet, in that case, would you not wind up with robotic sentences that could not actually be read as narrative.
Second: literal meaning should be retained whenever possible. That is not easy either, but any attempt at extracting data from narrative needs to be faithful to the meaning of the “story.” Literalness is not an easy criterion, since an isolated element from a story has far less likelihood of communicating original meaning than does the narrative sentence or phrase, from which it was drawn.
Third: distortion should be guarded against. That is, translation from narrative into data should not distort the meaning of the original communication, nor obscure its intent. Again, this is not easy to do. In the icteric patient noted above, the data should be able to be placed in a pre-designated pre-structured location on or in the chart. If no such data lacunae are present, say for “international travel,” an EHR should be able to create them, on the fly. And of course, guarding in reverse, suppose the patient’s hematocrit was 17 gm/L, extrapolation to the narrative should clearly say not simply “anemia,” but rather should indicate hemolysis on the basis of malaria.
In closing, consider again William Osler’s words. He said “Listen to the Patient.” He did not say “listen to the data.”
In medicine, we are admittedly in deep trouble. In the modern EHR realm we are in a related sort of trouble, since this tool of great promise subtly erects communicative walls between patient and provider, and we need to find ways of listening through. Listening means paying attention to the patient’s story, but neither exclusively to the narrative, nor solely to the elemental structured data that appear, but instead, to both at once.
Unless we recognize this need for harmonization, and unless we achieve the ability to create simple medical records with our ears, eyes, and minds attentive to the patient, our situation as providers will not improve. The individual who seeks our help must remain first in our minds, and thus, first in our electronic tools.
How we, as providers, recognize and deal with the “great divide” between story and data, how we select electronic tools that support our efforts to do so, is long going to be a key part of listening to the patient.
Xpress Technologies Classic and EHR software solutions were design to help the provider listen to the patient while documenting the story from their point of view in the simplest manner possible. Our solutions allow the provider to create a narrative while collecting key data points. We give providers the tools to provide excellent patient care while electronically recording the encounter.
- Dr. Donald Kamens, MD FACEP, FAAEM
Published November 2013
Humpty Dumpty sat on a wall,
Humpty Dumpty had a great fall.
All the king's horses and all the king's men
Couldn't put Humpty together again.
"I don't know what you mean by 'glory,' " Alice said.
Humpty Dumpty smiled contemptuously. "Of course you don't—till I tell you. I meant 'there's a nice knock-down argument for you!' "
"But 'glory' doesn't mean 'a nice knock-down argument'," Alice objected.
"When I use a word," Humpty Dumpty said, in rather a scornful tone, "it means just what I choose it to mean—neither more nor less."
"The question is," said Alice, "whether you can make words mean so many different things."
"The question is," said Humpty Dumpty, "which is to be master—that's all."